Opened 11 months ago
Closed 11 months ago
#71870 closed defect (fixed)
llama.cpp: builds fail for 10.13 and earlier, due to errant use of OpenMP
| Reported by: | mascguy (Christopher Nielsen) | Owned by: | i0ntempest |
|---|---|---|---|
| Priority: | Normal | Milestone: | |
| Component: | ports | Version: | 2.10.5 |
| Keywords: | Cc: | ||
| Port: | llama.cpp |
Description
While it's not clear whether this is an upstream issue per se - ideally the CMake build should auto-disable OpenMP support if not available - it appears that isn't the case. Instead, the build continues forward, and fails at link time due to missing symbols:
Undefined symbols for architecture x86_64:
"___kmpc_barrier", referenced from:
_ggml_barrier in ggml-cpu.c.o
_.omp_outlined. in ggml-cpu.c.o
"___kmpc_end_single", referenced from:
_.omp_outlined. in ggml-cpu.c.o
"___kmpc_fork_call", referenced from:
_ggml_graph_compute in ggml-cpu.c.o
"___kmpc_global_thread_num", referenced from:
_ggml_barrier in ggml-cpu.c.o
_ggml_graph_compute in ggml-cpu.c.o
"___kmpc_push_num_threads", referenced from:
_ggml_graph_compute in ggml-cpu.c.o
"___kmpc_single", referenced from:
_.omp_outlined. in ggml-cpu.c.o
"_omp_get_num_threads", referenced from:
_.omp_outlined. in ggml-cpu.c.o
"_omp_get_thread_num", referenced from:
_.omp_outlined. in ggml-cpu.c.o
But explicitly disabling it with the addition of the following configure arg, fixes the issue:
-DGGML_OPENMP=OFF
Change History (2)
comment:1 Changed 11 months ago by mascguy (Christopher Nielsen)
comment:2 Changed 11 months ago by Christopher Nielsen <mascguy@…>
| Resolution: | → fixed |
|---|---|
| Status: | assigned → closed |
Note: See
TracTickets for help on using
tickets.

PR created with fix: PR 27409 - llama.cpp: explicitly disable OpenMP, to fix builds for 10.13 and earlier